Vehicle Localization with Adaptive Covariance Noise Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing localization methods for motor vehicles require three-dimensional data from sensors like LIDAR or suffer from large variances in position estimation using monocular cameras, leading to implausible position changes and delayed noise compensation.
Innovation Solution
A processor circuit estimates vehicle position using sensor data from a monocular camera, forming feature data with map data, and employs a statistical observer model to adaptively model measurement noise through a covariance matrix, refining the Kalman filter to stabilize position estimation by fusing movement models and image evaluations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a Kalman filter is used to compensate for estimation variances in position estimation, then the position estimation becomes consistent over time, but the compensation is delayed because the filter must first estimate measurement noise across multiple individual images using recursive average
Solution Approach 1:
The patent applies preliminary action by pre-defining the covariance matrix of measurement noise based on characteristics of the environment sensor and map data, rather than estimating it recursively during operation. This allows the Kalman filter to immediately use accurate noise compensation without the delay of iterative estimation, while still achieving consistent position estimation over time
Solution Approach 2:
The patent makes the covariance matrix adaptive by defining it as a function of environmental factors such as lighting conditions, landmark density, and sensor characteristics. This dynamic adjustment allows the system to optimize noise compensation for current conditions without requiring delayed recursive estimation, resolving the contradiction between reliability and time loss
2Device complexity
If a monocular camera is used for localization, then the device complexity is reduced compared to LIDAR, but the measurement precision deteriorates due to lacking depth information
Solution Approach 1:
The patent introduces map data as an intermediary that provides the missing depth and spatial information. By matching sensor data from the monocular camera with pre-stored map data containing three-dimensional landmark positions, the system achieves accurate position estimation without requiring complex 3D sensors like LIDAR
Solution Approach 2:
The patent transforms the two-dimensional image data from the monocular camera into three-dimensional position information by utilizing the known three-dimensional coordinates of landmarks from map data. This parameter transformation allows the simple monocular camera to achieve measurement precision comparable to complex 3D sensing systems
3Productivity
If individual camera images are used for position estimation, then the productivity is increased by providing frequent position updates, but the reliability deteriorates due to variance and measurement noise causing implausible position switches
Solution Approach 1:
The patent implements feedback by using the Kalman filter to continuously compare new position estimates with previous estimates and predicted vehicle motion. The filter provides feedback correction that eliminates implausible position switches while maintaining high update frequency, resolving the contradiction between productivity and reliability
Data Source
AI summary
A method for localizing a motor vehicle in an environment during a driving operation comprises, by a processor circuit in repeated estimation cycles, receiving sensor data of landmarks of the environment from an environment sensor, and ascertaining a respective estimated position of the motor vehicle from feature data, which are formed of map data of a map region of the environment and of the sensor data, using an estimation module. A movement path is estimated from the positions of multiple of the estimation cycles using a statistical observer model, and the observer model, during the formation of the movement path, models measurement noise contained in the position data as a covariance matrix of position coordinates of the position data, with matrix values of the covariance matrix being ascertained by the estimation module as a function of the sensor data and the map data.